Cat oral health monitoring system based on multi-modal perception and ai assistance

Abstract

The application belongs to the field of pet health monitoring, and particularly relates to a cat oral cavity health monitoring system based on multi-modal sensing and AI assistance. The system comprises an intelligent interaction terminal and a cloud intelligent analysis platform; the terminal collects sulfide concentration, occlusion waveform and motion data through a multi-modal sensing module, and is provided with a labyrinth airway for preventing saliva pollution; the cloud platform constructs a personalized health baseline based on a long short-term memory self-encoder, and realizes abnormal detection and risk assessment. Through the above scheme, the application realizes non-invasive health monitoring and disease classification in the natural biting process of a cat, utilizes texture guide ridges to synchronously clean the periodontal friction, effectively solves the problems of early symptoms of oral diseases being hidden and domestic cats having low cleaning compliance, and realizes the change from passive medical treatment to active health management.

Description

Technical Field

[0001] This invention belongs to the field of pet health monitoring, specifically relating to a cat oral health monitoring system based on multimodal perception and AI assistance. Background Technology

[0002] With the acceleration of global urbanization and the deep development of the companion economy, ensuring the quality of life for family pets has become a key research focus in the field of pet healthcare. In home environments where cats are the primary companion animals, pet owners' parenting concepts are undergoing a paradigm shift from passive medical care to proactive health management. This shift has driven the development of non-invasive, intelligent daily monitoring technologies for pets' physiological states. Especially against the backdrop of increasing emphasis on pet welfare, building a preventative health early warning system that covers the entire life cycle of pets has become a core evolutionary direction in the current smart pet healthcare field.

[0003] In the field of pet health management, feline oral health has become a primary concern for maintaining the health of domestic cats due to its extremely high incidence and the insidious nature of its pathological evolution. Current oral care techniques typically rely on manual brushing or physical friction toys, aiming to mechanically remove plaque and prevent its mineralization into tartar. Meanwhile, clinical diagnosis primarily relies on veterinarians' visual examination and radiographic imaging to assess gingival congestion and alveolar bone resorption, determining whether the cat is in the pathological stage of gingivitis or periodontitis.

[0004] However, due to the strong pain tolerance instincts developed by felines during evolution, early oral lesions are often not readily apparent in external behavior. This physiological limitation means that existing visual observation methods are often constrained by the observer's subjective experience, leading to a significant delay in the diagnostic window. Patients are often only brought to the vet when they exhibit severe symptoms such as obvious halitosis or refusal to eat, resulting in irreversible periodontal tissue damage. Furthermore, the natural aversion to oral touch in domesticated cats means that forced artificial care not only induces severe stress but also makes it difficult to collect high-frequency, continuous health data. While current mechanical teeth-grinding products provide some physical friction, they are essentially a black box in terms of data feedback, unable to capture core biochemical indicators reflecting the degree of periodontal inflammation, such as volatile sulfide concentrations. Limited by the lack of perception of occlusal biomechanical characteristics, pet owners cannot perceive their cats' avoidance pain feedback during chewing, creating a monitoring vacuum for the dynamic trends of individualized oral lesions. Ultimately, this results in clinical intervention being unable to proceed in the early stages of disease due to the lack of a quantitative baseline. Summary of the Invention

[0005] This application provides a cat oral health monitoring system based on multimodal perception and AI assistance, which aims to solve the problems of hidden early symptoms of oral diseases in domestic cats leading to a delayed diagnosis and treatment window and low compliance with existing mechanical cleaning methods.

[0006] This application provides a first cat oral health monitoring system based on multimodal perception and AI assistance, including: an intelligent interactive terminal and a cloud-based intelligent analysis platform wirelessly connected to the intelligent interactive terminal; The intelligent interactive terminal includes a shell assembly, a multimodal sensing module, an edge computing module, and a wireless communication module. The multimodal sensing module includes a metal oxide semiconductor gas sensor, a pressure matrix sensing array, and an inertial measurement unit. The metal oxide semiconductor gas sensor is used to collect concentration data of volatile sulfides in the cavity when the intelligent interactive terminal is squeezed. The pressure matrix sensing array is used to record high-frequency dynamic bite waveform data when a cat bites the intelligent interactive terminal. The inertial measurement unit is used to acquire changes in the acceleration and angular velocity of the intelligent interactive terminal to determine the behavior trigger state. The cloud-based intelligent analysis platform includes a data fusion layer, an anomaly detection layer, and a risk assessment layer. The data fusion layer uses a 2-channel feature extraction network to process the concentration data and the dynamic bite waveform data respectively and generate a fused feature vector. The anomaly detection layer uses a long short-term memory autoencoder to perform reconstruction error analysis based on the personalized health baseline of a single cat, and determines that the reconstruction error exceeds a preset threshold as a physiological abnormality. The risk assessment layer uses an ensemble learning model to classify the physiological abnormalities as diseases.

[0007] In one embodiment of the present invention, the outer shell assembly adopts a layered structure of soft shell and hard core, and its exterior is provided with a clover-shaped structure adapted to the scissor bite characteristics of felines, and textured guide ridges are distributed on the end surface of the clover-shaped structure; the height and spacing of the textured guide ridges are set according to the gingival sulcus depth of an adult cat, and are used to mechanically clean the periodontal tissues during the biting interaction and guide the teeth into the preset detection area.

[0008] Furthermore, the intelligent interactive terminal is equipped with a labyrinth-style airflow channel, and the entrance of the labyrinth-style airflow channel is covered with an expanded polytetrafluoroethylene hydrophobic and breathable membrane; the expanded polytetrafluoroethylene hydrophobic and breathable membrane is used to block liquid saliva from entering the interior of the intelligent interactive terminal and allow gaseous volatile sulfur molecules to pass through, thereby achieving gas-liquid separation protection.

[0009] As one embodiment of the present invention, the intelligent interactive terminal is further provided with an independent olfactory induction chamber, which is used to place olfactory induction substances with biological preferences; the olfactory induction substances include catnip or knotweed, which induce cats to produce active biting behavior by releasing lactone substances, thereby ensuring the continuity of data collection.

[0010] Furthermore, the two-channel feature extraction network includes a 1D convolutional neural network channel and a recurrent neural network channel; the 1D convolutional neural network channel is used to extract the mechanical morphological features of the dynamic bite waveform, including peak pressure, rising edge slope, and pressure duration; the recurrent neural network channel is used to extract the temporal evolution features of volatile sulfide concentration.

[0011] In one embodiment of the present invention, the risk assessment layer adopts a limit gradient boosting tree model, and a Shaplega and interpretation model is connected after the risk assessment layer; the Shaplega and interpretation model calculates the marginal contribution of the biomechanical features and the temporal evolution features to the diagnostic results, transforms the disease classification results into a quantifiable feature attribution report, and pushes nursing suggestions to the user terminal based on the feature attribution report.

[0012] Furthermore, the anomaly detection layer performs the following steps during operation: acquiring health status data of a single cat within a preset period as a training set; using the training set to train the long short-term memory autoencoder to learn the physiological baseline of the specific individual; calculating the Euclidean distance between the input and output of the real-time acquired data after processing by the long short-term memory autoencoder; and triggering a health warning when the Euclidean distance continuously deviates from the distribution range of the physiological baseline.

[0013] As one embodiment of the present invention, the edge computing module of the intelligent interactive terminal integrates a behavior gating algorithm; the behavior gating algorithm uses motion data collected by the inertial measurement unit to identify active chewing behavior, static retrieval behavior, and idle state; when the identification result is the active chewing behavior, the edge computing module increases the sampling frequency of the pressure matrix sensing array and the metal oxide semiconductor gas sensor; when the identification result is the static retrieval behavior or the idle state, the edge computing module enters a low-power sleep mode.

[0014] Furthermore, the intelligent assistance platform utilizes transfer learning technology to load pre-trained feature extraction parameters on a public large-scale human activity recognition dataset, and performs fine-tuning with a small number of feline oral pathology labeled samples to improve the system's generalization performance and diagnostic accuracy in small sample environments.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention integrates volatile sulfur compound detection and occlusal dynamics analysis into a bionic terminal, achieving non-invasive monitoring based on a cat's natural biting instinct. This eliminates the stress response to cats caused by traditional oral examinations and solves the problem of medical delays caused by delayed perception by pet owners. The system's labyrinthine airflow channel and hydrophobic breathable membrane structure ensure ppb-level gas detection sensitivity while effectively avoiding damage to precision components from saliva contamination. A personalized health baseline is constructed through a long short-term memory autoencoder, enabling precise calibration of differences between individuals and overcoming the limited accuracy of general indicators in the field of veterinary medicine. Furthermore, the textured guide ridge design of the shell provides physical friction during monitoring, not only inhibiting plaque mineralization but also achieving a paradigm shift from passive medical treatment to proactive health management through hardware and software synergy, significantly reducing the frequency of subsequent complex surgeries. Attached Figure Description

[0016] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof.

[0017] Figure 1 This is a schematic diagram of the overall technical solution architecture of the cat oral health monitoring system based on multimodal perception and AI assistance proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of multimodal feature fusion and personalized anomaly detection in this invention; Figure 3 This is a schematic diagram of the multi-level interaction relationship and data flow between the intelligent interactive terminal and the cloud-based intelligent analysis platform in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In the accompanying drawings, the size of constituent elements, the thickness of layers, or areas may sometimes be exaggerated for clarity. Therefore, any implementation of this disclosure is not necessarily limited to the dimensions shown in the drawings, and the shapes and sizes of the components in the drawings do not reflect true proportions. Furthermore, the drawings schematically illustrate ideal examples, and any implementation of this disclosure is not limited to the shapes or values ​​shown in the drawings.

[0020] Reference Figure 1 , Figure 1 This is a schematic diagram of the overall technical architecture of the cat oral health monitoring system based on multimodal perception and AI assistance proposed in this invention. Figure 1 As shown, the system includes: an intelligent interactive terminal 100 and a cloud-based intelligent analysis platform 200 wirelessly connected to the intelligent interactive terminal 100.

[0021] The intelligent interactive terminal 100 serves as the hardware entity for data sensing and preprocessing, capturing physiological indicators of the cat's oral cavity through a dual mode of physical contact and chemical sensing. The intelligent interactive terminal 100 includes a shell assembly 110, a multimodal sensing module 120, an edge computing module 130, and a wireless communication module 140. The multimodal sensing module 120 is deployed within the shell assembly 110 at specific mechanical support positions and along airflow paths, achieving raw acquisition of physiological signals through hardware-level collaborative operation.

[0022] In this embodiment, during the biting interaction between the cat and the intelligent interactive terminal 100, the multimodal sensing module 120 of the intelligent interactive terminal 100 simultaneously activates multiple detection dimensions: When the terminal is squeezed, a metal oxide semiconductor gas sensor collects the concentration data of volatile sulfur compounds (VSCs) in the chamber through a specific airflow guidance mechanism; simultaneously, a pressure matrix sensing array records high-frequency dynamic biting waveform data during the cat's biting; furthermore, an inertial measurement unit (IMU) acquires the terminal's six-axis motion attitude information in real time, i.e., changes in acceleration and angular velocity, to determine whether the current interactive behavior meets a preset trigger threshold. The edge computing module 130 performs preliminary cleaning and packaging of the raw data, and sends the data stream to the cloud intelligent analysis platform 200 through the wireless communication module 140 (such as WiFi 6 or Bluetooth 5.2 protocol).

[0023] The cloud-based intelligent analysis platform 200 includes a data fusion layer 210, an anomaly detection layer 220, and a risk assessment layer 230. The data fusion layer 210 uses a two-channel feature extraction network to perform dimensionality reduction and feature reconstruction on the received multidimensional data; the anomaly detection layer 220 uses a Long Short-Term Memory Autoencoder (LSTM Autoencoder) to perform personalized baseline comparison to determine the offset of the current physiological state; and the risk assessment layer 230 uses an ensemble learning model to perform pathological classification on the identified anomalies. This application overcomes the limitations of single-dimensional monitoring being susceptible to interference through deep fusion of multimodal data, achieving accurate and non-invasive monitoring of the oral health status of cats.

[0024] Reference Figure 2 , Figure 2This is a schematic diagram illustrating the core principle framework of multimodal feature fusion and personalized anomaly detection in this invention. In this embodiment, the outer shell assembly 110 adopts a layered structure of soft shell and hard core. The outer shell assembly 110 has a clover-shaped structure adapted to the scissor bite characteristics of felines, and textured guide ridges are distributed on the end surface of this structure. The height and spacing of the textured guide ridges are strictly set at the micrometer level according to the gingival sulcus depth (0.5mm-1.0mm) of adult cats. When the cat makes a biting motion, the textured guide ridges clean the periodontal tissue through mechanical friction and guide the teeth into the preset detection area, ensuring that the pressure matrix sensing array can capture standardized occlusal features.

[0025] To ensure the accuracy of odor detection, the intelligent interactive terminal 100 is equipped with a labyrinthine airflow channel. The entrance to this channel is covered with a hydrophobic and breathable expanded polytetrafluoroethylene (ePTFE) membrane. When a cat chews, the gas generated by pressure in its mouth is forced into the labyrinthine channel. The ePTFE membrane, with its microporous structure, blocks the entry of liquid saliva while allowing gaseous volatile sulfur molecules to pass through efficiently, achieving gas-liquid separation and protecting the electrochemical stability of the internal precision sensors. Furthermore, the terminal contains an independent olfactory induction chamber containing olfactory inducing substances such as catnip or catnip, which release lactone compounds to induce a sustained chewing urge in the cat.

[0026] At the algorithm implementation level, the two-channel feature extraction network includes a 1D convolutional neural network (1D-CNN) channel and a recurrent neural network (RNN) channel. The 1D-CNN channel uses convolutional kernels to perform sliding window processing on the dynamic bite waveform to extract mechanical morphological features, including but not limited to peak pressure. Rising slope and duration of pressure : The RNN channels extract the temporal evolution features of volatile sulfur compound concentration through the temporal propagation of hidden states. The data fusion layer will and Concatenate into a global feature vector .

[0027] During execution, the anomaly detection layer 220 first acquires the health status data of a single cat within a preset period (e.g., 14 days) as a training set to train an LSTM autoencoder to learn the physiological baseline of that specific individual. During real-time monitoring, it calculates the real-time data sequence. Input and reconstructed output after autoencoder processing Euclidean distance between : When Euclidean distance When the confidence level consistently exceeds a preset threshold, the system determines it as a physiological abnormality and transmits it to the risk assessment layer 230. The risk assessment layer 230 uses the Extreme Gradient Boosting Tree (XGBoost) model for disease classification and connects the Shapley Tree and Explanation Model (SHAP) to calculate the marginal contribution of each feature, generating a feature attribution report.

[0028] Reference Figure 3 , Figure 3 This is a schematic diagram illustrating the multi-level interaction relationship and data flow between the intelligent interactive terminal and the cloud-based intelligent analysis platform in this invention. The edge computing module 130 of the intelligent interactive terminal 100 integrates a behavior gating algorithm, which utilizes motion data collected by the IMU. and The current status is classified in real time.

[0029] Specifically, the behavior gating algorithm identifies the following states through a classifier: active chewing behavior. Static retrieval behavior and vacant status When the recognition result is At that time, the edge computing module 130 executes a frequency compensation strategy, changing the sampling frequency of the pressure matrix sensing array from the base frequency. Upgrade to high-frequency sampling state (e.g., 500Hz), simultaneously increase the sampling frequency of the gas sensor; when the identification result is or At this time, the edge computing module 130 issues an interrupt command, causing the sensing module to enter a low-power sleep mode. This timing-triggered logic significantly extends the terminal's battery life and ensures the data granularity of high-value samples.

[0030] At the cloud-based learning level, the platform utilizes transfer learning technology to load general feature extraction weights pre-trained on a large-scale human activity recognition (HAR) dataset. Subsequently, by combining a small sample set with annotations of feline oral pathology (such as gingivitis and periodontal abscess), and by fine-tuning the hierarchical parameters, the model still has extremely strong generalization performance in the "small sample" environment of specific individuals, effectively solving the training overfitting problem caused by the scarcity of pet medical annotation data.

[0031] The cat oral health monitoring system based on multimodal perception and AI assistance proposed in this application has the following significant advantages: 1. Non-invasive interaction: By utilizing the cat's biting instinct through biomimetic design, physiological monitoring is achieved in a stress-free state, solving the problem of cats' high non-cooperation in traditional diagnosis and treatment methods.

[0032] 2. Completeness of detection dimensions: By coupling odor chemical characteristics with mechanical and dynamic characteristics, it can distinguish between physiological halitosis and malocclusion caused by pathological pain, with detection sensitivity reaching the ppb level.

[0033] 3. System stability and reliability: The physical barrier of the labyrinth airway and the ePTFE membrane effectively isolates the sensor from saliva contamination, ensuring the service life of the device in complex saliva environments.

[0034] 4. Personalized diagnostic accuracy: Individual baseline modeling based on LSTM autoencoder eliminates individual differences between cats of different breeds and ages, achieving true "one cat, one policy".

[0035] 5. Closed-loop management capability: By using the SHAP interpretation model, complex AI inferences are transformed into feature attribution reports that users can understand, realizing a closed loop of proactive health management from monitoring and early warning to care recommendations.

[0036] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. System embodiments are largely similar to method embodiments, and therefore are described more simply; relevant parts can be found in the descriptions of the method embodiments.

[0037] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, embodiments of this application can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of this application can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0038] The above provides a detailed description of a cat oral health monitoring system based on multimodal perception and AI assistance provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A cat oral health monitoring system based on multimodal perception and artificial intelligence assistance, characterized in that, include: The intelligent interactive terminal and the cloud-based intelligent analysis platform wirelessly connected to the intelligent interactive terminal; The intelligent interactive terminal includes a shell assembly, a multimodal sensing module, an edge computing module, and a wireless communication module. The multimodal sensing module includes a metal oxide semiconductor gas sensor, a pressure matrix sensing array, and an inertial measurement unit. The cloud-based intelligent analysis platform includes a data fusion layer, an anomaly detection layer, and a risk assessment layer. The inertial measurement unit is used to acquire the acceleration and angular velocity changes of the intelligent interactive terminal to determine the behavior trigger state. The metal oxide semiconductor gas sensor is used to collect the concentration data of volatile sulfur compounds in the cavity when the intelligent interactive terminal is squeezed. The pressure matrix sensing array is used to record high-frequency dynamic bite waveform data when biting behavior is detected. The data fusion layer uses a 2-channel feature extraction network to process the concentration data and the dynamic bite waveform data respectively and generate a fused feature vector. The anomaly detection layer uses a long short-term memory autoencoder to perform reconstruction error analysis based on the personalized health baseline of a single cat, and determines a physiological abnormality when the reconstruction error exceeds a preset threshold. The risk assessment layer uses an ensemble learning model to classify the physiological abnormalities into diseases.

2. The cat oral health monitoring system based on multimodal perception and artificial intelligence assistance according to claim 1, characterized in that, The outer shell assembly adopts a layered structure of soft shell and hard core, and has a clover-shaped structure on its exterior. Textured guide ridges are distributed on the end surface of the clover-shaped structure. The height and spacing of the textured guide ridges are set according to the gingival sulcus depth of an adult cat.

3. The cat oral health monitoring system based on multimodal perception and artificial intelligence assistance according to claim 1, characterized in that, The intelligent interactive terminal has a maze-like airflow channel inside, and the entrance of the maze-like airflow channel is covered with an expanded polytetrafluoroethylene hydrophobic and breathable membrane; the expanded polytetrafluoroethylene hydrophobic and breathable membrane is used to block liquid saliva from entering the interior of the intelligent interactive terminal and allow gaseous volatile sulfur molecules to pass through.

4. The cat oral health monitoring system based on multimodal perception and artificial intelligence assistance according to claim 1, characterized in that, The intelligent interactive terminal also has an independent olfactory induction chamber inside, which is used to place olfactory induction substances with biological preferences.

5. The cat oral health monitoring system based on multimodal perception and artificial intelligence assistance according to claim 1, characterized in that, The two-channel feature extraction network includes a 1D convolutional neural network channel and a recurrent neural network channel; the 1D convolutional neural network channel is used to extract the mechanical morphological features of the dynamic bite waveform, including peak pressure, rising edge slope, and pressure duration. The recurrent neural network channel is used to extract the temporal evolution features of the concentration data.

6. The cat oral health monitoring system based on multimodal perception and artificial intelligence assistance according to claim 5, characterized in that, The risk assessment layer adopts a limit gradient boosting tree model, and a Shaplega and interpretation model is connected after the risk assessment layer; the Shaplega and interpretation model converts the disease classification results into feature attribution reports by calculating the marginal contribution of the mechanical morphological features and the temporal evolution features to the diagnostic results.

7. The cat oral health monitoring system based on multimodal perception and artificial intelligence assistance according to claim 1, characterized in that, The anomaly detection layer performs the following steps during operation: acquiring health status data of a single cat within a preset period as a training set; and using the training set to train the long short-term memory autoencoder to learn the physiological baseline of that specific individual. Calculate the Euclidean distance between the input and output of the real-time acquired data after processing by the long short-term memory autoencoder; trigger a health warning when the Euclidean distance continuously deviates from the distribution range of the physiological baseline.

8. The cat oral health monitoring system based on multimodal perception and artificial intelligence assistance according to claim 1, characterized in that, The edge computing module integrates a behavior gating algorithm; the behavior gating algorithm uses motion data collected by the inertial measurement unit to identify active chewing behavior, static picking behavior, and idle state.

9. The cat oral health monitoring system based on multimodal perception and artificial intelligence assistance according to claim 8, characterized in that, When the identification result is the active chewing behavior, the edge computing module increases the sampling frequency of the pressure matrix sensing array and the metal oxide semiconductor gas sensor; when the identification result is the static picking behavior or the idle state, the edge computing module enters a low-power sleep mode.

10. The cat oral health monitoring system based on multimodal perception and artificial intelligence assistance according to claim 1, characterized in that, The cloud-based intelligent analysis platform utilizes transfer learning technology to load pre-trained feature extraction parameters on a public large-scale human activity recognition dataset, and then fine-tunes them by combining them with feline oral pathology annotation samples.

Images